New text-to-speech engine from AWS sounds like newsreader

Amazon Web Services (AWS) has launched a number of new neural text-to-speech models. One of those models has a style that has to imitate the way the newsreader sounds.

AWS states that speech quality is important, but that more can be done to make a synthetic voice sound more realistic, says TechCrunch.

Human ears can recognise the difference between, for example, a news broadcast, a sports broadcast and a lesson at school. Most people adopt the right speech style for the right context, and this helps to get the message across, according to the company.

Amazon Polly Newscaster

The new newsreading style, called Amazon Polly Newscaster, is now available as two American voices. It’s a male voice called Matthew and a female voice called Joanna. The voices are already being used by USA Today and the Canadian company The Globe and Mail to read lyrics.

According to Amazon, Polly Newscaster makes speech more realistic in a news broadcast. The voice must sound like what a user expects from a TV or radio broadcast.

With Amazon Translate, broadcasts can also be automatically translated into a language that the user understands.

Neural Text-to-Speech

Amazon Polly, the service that converts text to speech, also has a new feature called Neural Text-to-Speech (NTTS). NTTS is the result of years of Amazon research into text-to-speech and the use of machine learning.

According to Amazon, NTTS itself delivers significant improvements in speech quality. This makes the spoken text more natural and expressive. The new engine contains eleven voices. Three are for British English (Amy, Emma and Brian) and eight speak American English – Ivy, Joanna, Kendra, Kimberly, Salli, Joey, Justin and Matthew.

The new features have been made directly available in US East (N. Virginia), US West (Oregon) and Europe (Ireland). A free subscription is available, offering 1 million characters per month for NTTS voices for the first twelve months.